Who is
reading?

What these numbers do not say.

Counts are server-side HTTP requests. They are not people, sessions or visits.

Agent identity is a header the client chose to send. Where a provider publishes the addresses its agents use, the claim is checked against that published list and reported as verified; where no list is published it is reported as declared and nothing more.

A claim contradicted by its own provider's published addresses is recorded as refuted and excluded from every total on this page.

No address, cookie, session, fingerprint or query string is stored. The finest record kept is an hourly count per path, agent and outcome.

Paths this site does not publish are counted together as one line rather than named.

Figures are recomputed at most once every five minutes, so they trail live traffic by up to that long.

Verification uses a dated snapshot of each provider’s own published address list, retrieved 2026-09-07 and held in this repository: 646 address ranges across 8 published documents. Anthropic, OpenAI, Perplexity, Google and Microsoft publish one. Apple and most others do not, so their agents can be recorded and never confirmed.

Hourly counts are kept for 400 days and then deleted. Definitions for every category are published with the counts at /api/readership, and the classifier, the address snapshot and the retention rule are in the source of this site.